Autonomous oceanographic sampling networks (AOSNs) are groups of autonomous underwater vehicles and other instrument platforms. They were envisioned to address the serious undersampling problem that exists for the ocean. Advanced AOSNs will include numerous heterogeneous components and perform complex, long-duration missions. They will have to cope with vehicles failing, new vehicles arriving, and changes in the mission and environment. Controlling such a system is a very challenging task. The CoDA project focuses on developing intelligent control mechanisms for advanced AOSNs. CoDA treats AOSN components as agents that follow protocols to create an effective organization to carry out the mission. CoDA is a two-level approach. A metalevel organization (MLO) first self-organizes from a subset of the agents to discover the AOSN's total capability repertoire, then designs a task-level organization (TLO) to efficiently carry out the mission. When changes occur, the MLO can step in to reorganize the AOSN to fit the changed situation. This two-level approach allows the AOSN to be both efficient and flexible in the face of change. In addition to describing CoDA, this paper also presents simulation results to show its behavior, both normally and when agents fail. Index terms-Autonomous oceanographic sampling networks, autonomous underwater vehicles, multiagent systems.